-
Notifications
You must be signed in to change notification settings - Fork 35
/
collect_results.py
177 lines (135 loc) · 6.76 KB
/
collect_results.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
# collects all saved results from experiments on given datasets
#
# USAGE:
# python collect_results.py dataset1 dataset2 ...
import sys
import itertools
import numpy as np
import experiments as ex
import util
split = 'tst' # choose which data split to evaluate on 'trn', 'val' or 'tst'
n_err = 2 # number of stds in error bars
bits_per_pixel = False # whether to use bits/pixel instead of log likelihood (only for image datasets)
root_results = 'results/' # folder where to save results
def calc_bits_per_pixel(lp, lp_err):
Dlog2 = ex.data.n_dims * np.log(2)
bpp = -lp / Dlog2 + 8
bpp_err = lp_err / Dlog2
return bpp, bpp_err
def result(model_name, mode, n_hiddens, act_fun, n_comps=None, batch_norm=False):
try:
model = ex.load_model(model_name, mode, n_hiddens, act_fun, n_comps, batch_norm)
except IOError:
return 'N/A'
res, err = ex.evaluate_logprob(model, split, use_image_space=bits_per_pixel)
if bits_per_pixel:
res, err = calc_bits_per_pixel(res, err)
return '{0:.2f} +/- {1:.2f}'.format(res, n_err * err)
def collect_results(data, n_hiddens, n_layers, n_comps, n_layers_comps, act_funs, modes, has_cond):
print 'collecting for {0}...'.format(data)
ex.load_data(data)
# create file to write to
filename = ('{0}_{1}_bpp.txt' if bits_per_pixel else '{0}_{1}.txt').format(data, split)
util.make_folder(root_results)
f = open(root_results + filename, 'w')
f.write('Results for {0}\n'.format(data))
f.write('\n')
for act, mode in itertools.product(act_funs, modes):
f.write('actf: {0}\n'.format(act))
f.write('mode: {0}\n'.format(mode))
f.write('\n')
# gaussian
f.write('Gaussian\n')
res, err = ex.fit_and_evaluate_gaussian(split, cond=False, use_image_space=bits_per_pixel)
if bits_per_pixel:
res, err = calc_bits_per_pixel(res, err)
f.write(' {0:.2f} +/- {1:.2f}\n'.format(res, n_err * err))
if has_cond:
f.write('conditional\n')
res, err = ex.fit_and_evaluate_gaussian(split, cond=True, use_image_space=bits_per_pixel)
if bits_per_pixel:
res, err = calc_bits_per_pixel(res, err)
f.write(' {0:.2f} +/- {1:.2f}\n'.format(res, n_err * err))
f.write('\n')
# made
f.write('MADE 1 comp\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('made', mode, [nh]*1, act)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('made', mode, [nh]*2, act)))
if has_cond:
f.write('conditional\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('made_cond', mode, [nh]*1, act)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('made_cond', mode, [nh]*2, act)))
f.write('\n')
# mog made
for nc in n_comps:
f.write('MADE {0} comp\n'.format(nc))
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('made', mode, [nh]*1, act, nc)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('made', mode, [nh]*2, act, nc)))
if has_cond:
f.write('conditional\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('made_cond', mode, [nh]*1, act, nc)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('made_cond', mode, [nh]*2, act, nc)))
f.write('\n')
# real nvp
for nl in n_layers:
f.write('RealNVP {0} layers\n'.format(nl))
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('realnvp', None, [nh]*1, 'tanhrelu', nl, True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('realnvp', None, [nh]*2, 'tanhrelu', nl, True)))
if has_cond:
f.write('conditional\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('realnvp_cond', None, [nh]*1, 'tanhrelu', nl, True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('realnvp_cond', None, [nh]*2, 'tanhrelu', nl, True)))
f.write('\n')
# maf
for nl in n_layers:
f.write('MAF {0} layers\n'.format(nl))
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('maf', mode, [nh]*1, act, nl, True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('maf', mode, [nh]*2, act, nl, True)))
if has_cond:
f.write('conditional\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('maf_cond', mode, [nh]*1, act, nl, True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('maf_cond', mode, [nh]*2, act, nl, True)))
f.write('\n')
# maf on made
for nl, nc in n_layers_comps:
f.write('MAF {0} layers on MADE {1} comp\n'.format(nl, nc))
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('maf_on_made', mode, [nh]*1, act, [nl, nc], True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('maf_on_made', mode, [nh]*2, act, [nl, nc], True)))
if has_cond:
f.write('conditional\n')
for nh in n_hiddens:
f.write(' [1 x {0}]: {1}\n'.format(nh, result('maf_on_made_cond', mode, [nh]*1, act, [nl, nc], True)))
f.write(' [2 x {0}]: {1}\n'.format(nh, result('maf_on_made_cond', mode, [nh]*2, act, [nl, nc], True)))
f.write('\n')
# close file
f.close()
def main():
for data in sys.argv[1:]:
if data == 'power':
collect_results(data, [100], [5, 10], [10], [(5, 10)], ['relu'], ['sequential'], False)
elif data == 'gas':
collect_results(data, [100], [5, 10], [10], [(5, 10)], ['tanh'], ['sequential'], False)
elif data == 'hepmass':
collect_results(data, [512], [5, 10], [10], [(5, 10)], ['relu'], ['sequential'], False)
elif data == 'miniboone':
collect_results(data, [512], [5, 10], [10], [(5, 10)], ['relu'], ['sequential'], False)
elif data == 'bsds300':
collect_results(data, [512, 1024], [5, 10], [10], [(5, 10)], ['relu'], ['sequential'], False)
elif data == 'mnist':
collect_results(data, [1024], [5, 10], [10], [(5, 10)], ['relu'], ['sequential'], True)
elif data == 'cifar10':
collect_results(data, [1024, 2048], [5, 10], [10], [(5, 10)], ['relu'], ['random'], True)
else:
print '{0} is not a valid dataset'.format(data)
continue
if __name__ == '__main__':
main()